Fix NOT_COORDINATOR error caused by wrong brokers in Metadata response

[BewareMyPower]

Motivation

After #1107 was cherry-picked into branch-2.8.2, some tests of KafkaListenerNameTest are failed. Because that PR forgot to migrate the following code:

kop/kafka-impl/src/main/java/io/streamnative/pulsar/handlers/kop/KafkaRequestHandler.java

Lines 792 to 796 in 51563fe

	synchronized (allNodes) {
	if (!allNodes.stream().anyMatch(node1 -> node1.equals(newNode))) {
	allNodes.add(newNode);
	}
	}

After that when the advertised listener is different with the listener, the brokers field in MetadataResponse doesn't contain the advertised listener, while the leader and ISR in MetadataResponse.PartitionMetadata use the advertised listener. Then Kafka client won't be able to find the advertised listener in brokers field.

Modifications

Add the partition leader's node to the brokers field if the brokers field doesn't contain the node.

[BewareMyPower]

We should also investigate why the tests in master and branch-2.9.2 succeeded.

[BewareMyPower]

It looks like in master branch, as well as branch-2.9.2 and branch-2.10.0, the leader and ISR fields both use the listener (0.0.0.0:15003) but not the advertised listener (localhost:15003). This behavior has already been introduced before #1107.

However, in branch-2.8.2, the leader and ISR fields use the advertised listener (localhost:15003).